There are many situations that can block a proxy, including writing to syslog, dns and DB queries,... At the end you cannot rely on a single app to have a smooth runtime, the ecosystem has to be also in good parameters.
The current implementation is like this now, the choice of developer. You can improve it if you want, or use the old no-db mode. My patch ensured that the old mode work as it was before adding db support. That was a regression. But what is offered with db support, is what the developer added, and apparently he did it for all instances to be available at that moment. Maybe a note about that can be added in the docs.
Otherwise similar case can be that sqlops (or db_cluster, not sure by heart) didn't start without all db servers available. Recently someone proposed a patch to allow sqlops even when connection fails. We had blocking tcp for years, it was not disabled because of that. So things can be improved as one needs something different, it is open source...
Cheers,
Daniel